MCPcopy
hub / github.com/llmware-ai/llmware / prune_context

Method prune_context

llmware/models.py:2926–2946  ·  view source on GitHub ↗
(self, ctx, front=100,back=100)

Source from the content-addressed store, hash-verified

2924 return {}
2925
2926 def prune_context(self, ctx, front=100,back=100):
2927
2928 # apply pruning of stop words
2929 pruned_ctx = Utilities().prune_stop_words(ctx,front=front,back=back)
2930
2931 # test len
2932 pruned_tokens = self.count_tokens(pruned_ctx)
2933
2934 logger.info(f"BaseModel - prune_context - token count - {pruned_tokens}")
2935
2936 # extra pruning for very large contexts
2937 # need to reduce for 14B parameter models
2938 if pruned_tokens > 16000:
2939 start = pruned_ctx[0:1000]
2940 end = pruned_ctx[pruned_tokens-5000:]
2941 super_pruned = start + end
2942 pruned_tokens = self.count_tokens(super_pruned)
2943 logger.info(f"BaseModel - prune_context - token count - {pruned_tokens}")
2944 pruned_ctx = super_pruned
2945
2946 return pruned_ctx
2947
2948 def count_tokens(self, ctx, tokenizer=None):
2949

Callers

nothing calls this directly

Calls 3

count_tokensMethod · 0.95
UtilitiesClass · 0.90
prune_stop_wordsMethod · 0.80

Tested by

no test coverage detected